Neel Nanda
Person on 4 sites
Neel Nanda runs the mechanistic interpretability team at Google DeepMind, which tries to reverse-engineer the algorithms trained neural networks learn.
-
5 facts on wiqqi (opens wiqqi.app)
- Google DeepMind: Runs the mechanistic interpretability team (as of November 2026) (opens wiqqi.app)
- Before: Language model interpretability researcher at Anthropic under Chris Olah; independent interpretability research (opens wiqqi.app)
- Education: Pure maths undergraduate, Cambridge, graduated 2020 (opens wiqqi.app)
-
3 channels on ppll (opens ppll.app)
-
x.com: @NeelNanda5 (opens x.com)
“Another good contribution to the "no, we have not solved interpretability" discourse”
-
YouTube: Neel Nanda (opens www.youtube.com)
“Don't Go Quiet On Me (by Opus 5.5)”
- Site: neelnanda.io (opens www.neelnanda.io)
-
x.com: @NeelNanda5 (opens x.com)
-
17 beliefs on korrents (opens korrents.com)
-
Complete reverse-engineering of neural networks is a less promising research direction than studying model biology and applying interpretability usefully (opens korrents.com)
“I'm now fairly pessimistic about ambitious interpretability (i.e. complete reverse-engineering), and I'm excited about model biology (studying qualitative high-level properties of models) and applied interpretability…”
-
Sparse autoencoders are a useful interpretability tool but are often used wastefully when a simpler, more obvious method would work as well or better (opens korrents.com)
“I'm more agnostic about the best techniques, things like sparse autoencoders are a useful tool, but easy to waste effort using when a simpler method is sufficient or better - start by doing the obvious thing!”
-
Research taste, the harder-to-learn conceptual skills behind good research strategy, takes a long time to develop but very little time to apply once acquired (opens korrents.com)
“My model is that research requires a mix of skills. The day-to-day coding and execution is crucial. But there's also a set of harder-to-learn conceptual skills, collectively called research taste. These skills take a…”
-
Complete reverse-engineering of neural networks is a less promising research direction than studying model biology and applying interpretability usefully (opens korrents.com)
-
8 recommendations on rcmnd (opens rcmnd.app)
-
3Blue1Brown (Essence of Linear Algebra) (opens rcmnd.app)
“Linear Algebra: 3Blue1Brown or Linear Algebra Done Right”
-
Think Python 2e (opens rcmnd.app)
“If you prefer a more traditional textbook, Think Python 2e is excellent and also available freely online.”
-
einops (opens rcmnd.app)
“I highly, highly recommend learning how to use einops (a library to nicely do any reasonable manipulation of a single tensor) and einsum (a built in torch function implementing Einstein Summation notation, to do…”
-
3Blue1Brown (Essence of Linear Algebra) (opens rcmnd.app)